MaxView

Timing Test Reference Updates
‹ 10_shardy_falseCase: 10_shardy_true11_optimizer_offload_false ›

Metrics: main (7f7822822) vs feat/nnx-post-train-fixes (2fbc8c2f1)

Metricmain  7f7822822feat/nnx-post-train-fixes  2fbc8c2f1Diff (feat/nnx-post-train-fixes − main)
Parameters1.104 billion1.104 billion—
Final loss8.17908.17900
TFLOP/s92.04390.827-1.216
Tok/s13873.713690.5-183.261
Avg s/step3.3582.719-0.639
Memory %1.381.380
JAX0.9.20.9.2—

Diff = branch value − main value. Green = branch improved. Red = branch regressed.

main  ·  7f7822822  ·  main_20260417_125726  ·  full log 
XPK Start: Fri Apr 17 13:51:33 UTC 2026
PyTorch was not found. Models won't be available and only tokenizers, configuration and file/data utilities can be used.
2026-04-17 13:51:57.840192: E external/local_xla/xla/stream_executor/cuda/cuda_platform.cc:51] failed call to cuInit: INTERNAL: CUDA error: Failed call to cuInit: UNKNOWN ERROR (303)
I0417 13:51:58.022227 138211156707136 max_utils.py:273] Attempting to initialize the jax distributed system...
I0417 13:52:07.064029 138211156707136 distributed.py:149] Starting JAX distributed service on [::]:8482
I0417 13:52:07.066258 138211156707136 distributed.py:172] Connecting to JAX distributed service on mt-10-shardy-true-qoo5k-slice-job-0-0.mt-10-shardy-true-qoo5k:8482
I0417 13:52:08.317111 138211156707136 max_utils.py:284] Jax distributed system initialized!
I0417 13:52:14.378309 138211156707136 max_utils.py:800] System Information: Jax Version: 0.9.2
I0417 13:52:14.378410 138211156707136 max_utils.py:801] System Information: Jaxlib Version: 0.9.2
I0417 13:52:14.378450 138211156707136 max_utils.py:802] System Information: Jax Backend: PJRT C API
TFRT TPU v6 lite
Built on Mar 4 2026 11:32:08 (1772652728) cl/878335365
I0417 13:52:14.378485 138211156707136 train_utils.py:348] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing.
I0417 13:52:15.399107 138211156707136 maxtext_utils.py:1548] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1)
I0417 13:52:15.399384 138211156707136 checkpointing.py:677] Setting up checkpoint logger...
I0417 13:52:15.399436 138211156707136 checkpointing.py:233] Creating checkpoint manager with ocdbt=True and zarr3=True
I0417 13:52:15.399480 138211156707136 pytree_checkpoint_handler.py:592] save_device_host_concurrent_bytes=None
I0417 13:52:15.399815 138211156707136 base_pytree_checkpoint_handler.py:441] Created BasePyTreeCheckpointHandler: use_ocdbt=True, use_zarr3=True, pytree_metadata_options=PyTreeMetadataOptions(support_rich_types=False), array_metadata_store=<orbax.checkpoint._src.metadata.array_metadata_store.Store object at 0x7db336f18980>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB)
I0417 13:52:24.041121 138211156707136 checkpointing.py:265] Enabling policy for fixed interval checkpointing.
I0417 13:52:24.041350 138211156707136 checkpoint_manager.py:708] [process=0][thread=MainThread] CheckpointManager init: checkpointers=None, item_names=('items',), item_handlers={'items': <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7db2075ede80>}, handler_registry=None
I0417 13:52:24.041582 138211156707136 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7db2075ede80>` for item "items" and save args `<class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeSaveArgs'>` and restore args `<class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>` to `_handler_registry`.
I0417 13:52:24.041628 138211156707136 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7db207729a30>` for item "metrics" and save args `<class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>` and restore args `<class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>` to `_handler_registry`.
I0417 13:52:24.041663 138211156707136 composite_checkpoint_handler.py:505] Initialized registry DefaultCheckpointHandlerRegistry({('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeSaveArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7db2075ede80>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7db2075ede80>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7db207729a30>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7db207729a30>}).
I0417 13:52:24.042007 138211156707136 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.34
I0417 13:52:24.042076 138211156707136 async_checkpointer.py:192] [process=0][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>._fn at 0x7d9edc128180> timeout: 1200 secs and primary_host=0 for async checkpoint writes
I0417 13:52:26.279558 138211156707136 checkpoint_manager.py:1812] Found 0 checkpoint steps in gs://lance-maxtext/linen_ckpt_xpk_main_20260417_125726/linen_xpk_main_20260417_125726_10_shardy_true/checkpoints
I0417 13:52:26.653001 138211156707136 checkpoint_manager.py:929] [process=0][thread=MainThread] CheckpointManager created,  primary_host=0, CheckpointManagerOptions=CheckpointManagerOptions(save_interval_steps=1, max_to_keep=None, keep_time_interval=None, keep_period=None, should_keep_fn=None, best_fn=None, best_mode='max', keep_checkpoints_without_metrics=True, step_prefix=None, step_format_fixed_length=None, step_name_format=None, create=True, cleanup_tmp_directories=False, save_on_steps=frozenset(), single_host_load_and_broadcast=False, todelete_subdir=None, todelete_full_path=None, enable_background_delete=False, read_only=False, enable_async_checkpointing=True, async_options=None, multiprocessing_options=MultiprocessingOptions(primary_host=0, active_processes=None, barrier_sync_key_prefix=None), should_save_fn=None, file_options=FileOptions(path_permission_mode=None), save_root_metadata=True, temporary_path_class=None, save_decision_policy=FixedIntervalPolicy(interval=10), preservation_policy=LatestN(n=None), prevent_write_metrics=False, enable_should_save_is_saving_in_progress_check=True, enable_per_process_directory_creation=False, lightweight_initialize=False), root_directory=gs://lance-maxtext/linen_ckpt_xpk_main_20260417_125726/linen_xpk_main_20260417_125726_10_shardy_true/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7db2077b0f20>
I0417 13:52:26.653175 138211156707136 checkpointing.py:301] Checkpoint manager created!
I0417 13:52:27.573287 138211156707136 nnx_wrappers.py:437] Unknown Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed').
I0417 13:52:27.573395 138211156707136 nnx_wrappers.py:437] Unknown Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0417 13:52:27.949884 138211156707136 attentions.py:1088] attentions/inputs_q Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed').
I0417 13:52:27.949971 138211156707136 attentions.py:1088] attentions/inputs_q Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0417 13:52:27.966227 138211156707136 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed').
I0417 13:52:27.966284 138211156707136 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0417 13:52:27.990094 138211156707136 attentions.py:1154] attentions/query Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0417 13:52:27.990169 138211156707136 attentions.py:1154] attentions/query Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0417 13:52:28.006285 138211156707136 attentions.py:1155] attentions/key Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0417 13:52:28.006350 138211156707136 attentions.py:1155] attentions/key Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0417 13:52:28.022442 138211156707136 attentions.py:1156] attentions/value Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0417 13:52:28.022508 138211156707136 attentions.py:1156] attentions/value Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0417 13:52:28.048163 138211156707136 attentions.py:1197] attentions/out Logical: bfloat16[32,2048,16,128].................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv').
I0417 13:52:28.048236 138211156707136 attentions.py:1197] attentions/out Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0417 13:52:28.069182 138211156707136 linears.py:525] linears/x Logical: bfloat16[32,2048,7168]...................................... ('activation_batch', 'activation_length', 'activation_mlp').
I0417 13:52:28.069247 138211156707136 linears.py:525] linears/x Physical: bfloat16[32,2048,7168]...................................... ('fsdp', None, None).
I0417 13:52:28.286100 138211156707136 checkpointing.py:577] checkpoint manager exists so trying to load this run's existing checkpoint
I0417 13:52:28.286211 138211156707136 checkpointing.py:665] No existing checkpoints found, not restoring checkpoint.
fsdp: 32
I0417 13:52:29.698318 138211156707136 maxtext_utils.py:1651]  params/params/decoder/decoder_norm/scale
    Shape:     float32[2048]
    Logical:   P('norm',)
    Physical:  (None,)
I0417 13:52:29.698450 138211156707136 maxtext_utils.py:1651]  params/params/decoder/layers/mlp/wi_0/kernel
    Shape:     float32[2048,16,7168]
    Logical:   P('embed', 'layers', 'mlp')
    Physical:  ('fsdp', None, None)
I0417 13:52:29.698504 138211156707136 maxtext_utils.py:1651]  params/params/decoder/layers/mlp/wi_1/kernel
    Shape:     float32[2048,16,7168]
    Logical:   P('embed', 'layers', 'mlp')
    Physical:  ('fsdp', None, None)
I0417 13:52:29.698562 138211156707136 maxtext_utils.py:1651]  params/params/decoder/layers/mlp/wo/kernel
    Shape:     float32[7168,16,2048]
    Logical:   P('mlp', 'layers', 'embed')
    Physical:  (None, None, 'fsdp')
I0417 13:52:29.698613 138211156707136 maxtext_utils.py:1651]  params/params/decoder/layers/post_self_attention_layer_norm/scale
    Shape:     float32[2048,16]
    Logical:   P('norm', 'layers')
    Physical:  (None, None)
I0417 13:52:29.698650 138211156707136 maxtext_utils.py:1651]  params/params/decoder/layers/pre_self_attention_layer_norm/scale
    Shape:     float32[2048,16]
    Logical:   P('norm', 'layers')
    Physical:  (None, None)
I0417 13:52:29.698705 138211156707136 maxtext_utils.py:1651]  params/params/decoder/layers/self_attention/key/kernel
    Shape:     float32[2048,16,16,128]
    Logical:   P('embed', 'layers', 'kv_heads', 'kv_head_dim')
    Physical:  ('fsdp', None, None, None)
I0417 13:52:29.698755 138211156707136 maxtext_utils.py:1651]  params/params/decoder/layers/self_attention/out/kernel
    Shape:     float32[16,16,128,2048]
    Logical:   P('heads', 'layers', 'kv', 'embed')
    Physical:  (None, None, None, 'fsdp')
I0417 13:52:29.698792 138211156707136 maxtext_utils.py:1651]  params/params/decoder/layers/self_attention/query/kernel
    Shape:     float32[2048,16,16,128]
    Logical:   P('embed', 'layers', 'q_heads', 'kv')
    Physical:  ('fsdp', None, None, None)
I0417 13:52:29.698827 138211156707136 maxtext_utils.py:1651]  params/params/decoder/layers/self_attention/value/kernel
    Shape:     float32[2048,16,16,128]
    Logical:   P('embed', 'layers', 'kv_heads', 'kv_head_dim')
    Physical:  ('fsdp', None, None, None)
I0417 13:52:29.698888 138211156707136 maxtext_utils.py:1651]  params/params/decoder/logits_dense/kernel
    Shape:     float32[2048,32000]
    Logical:   P('embed_vocab', 'vocab')
    Physical:  ('fsdp', None)
I0417 13:52:29.698937 138211156707136 maxtext_utils.py:1651]  params/params/token_embedder/embedding
    Shape:     float32[32000,2048]
    Logical:   P('vocab', 'embed_vocab')
    Physical:  (None, 'fsdp')

I0417 13:52:30.183752 138211156707136 train.py:155] train/xent Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length').
I0417 13:52:30.183859 138211156707136 train.py:155] train/xent Physical: float32[32,2048]............................................ ('fsdp', None).
I0417 13:52:30.199181 138211156707136 train.py:162] train/z_loss Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length').
I0417 13:52:30.199241 138211156707136 train.py:162] train/z_loss Physical: float32[32,2048]............................................ ('fsdp', None).
I0417 13:52:41.156275 138211156707136 max_utils.py:791] Total memory size: 1.5 GB, Output size: 0.4 GB, Temp size: 1.1 GB, Argument size: 0.4 GB, Host temp size: 0.0 GB.
I0417 13:52:41.164126 138211156707136 metric_logger.py:301] number parameters: 1.104 billion
I0417 13:52:52.703328 138211156707136 checkpointing.py:772] Waiting for step 0 to finish before checkpoint...
I0417 13:52:52.856797 138211156707136 checkpointing.py:776] Waited 0.15344500541687012 seconds for step 0 to finish before starting checkpointing.
I0417 13:52:52.859305 138211156707136 checkpoint_manager.py:2009] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0417 13:52:52.861318 138211156707136 checkpoint_manager.py:1512] [process=0] Saving checkpoint at step 0
I0417 13:52:52.862731 138211156707136 event_tracking.py:70] [process=0] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_main_20260417_125726/linen_xpk_main_20260417_125726_10_shardy_true/checkpoints/0.
I0417 13:52:54.038592 138211156707136 signaling_client.py:364] Using JaxDistributedSignalingClient
I0417 13:52:54.052600 138211156707136 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array.
I0417 13:52:54.052665 138211156707136 replica_slices.py:424] Transferring arrays to host memory with options: use_replica_parallel=True, min_slice_bytes_for_replica_parallel=None, max_replicas_for_replica_parallel=None, enable_pinned_host_transfer=False
I0417 13:52:54.323514 138211156707136 base_pytree_checkpoint_handler.py:154] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.271608s
I0417 13:52:54.324312 138211156707136 base_pytree_checkpoint_handler.py:130] [process=0] /jax/orbax/write/blocking_gbytes_per_sec: 5.547 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.2781085968017578 s) (per-host)
I0417 13:52:54.324379 138211156707136 base_pytree_checkpoint_handler.py:768] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 0.278195s (batch_requests_ready=0.002336s, total_serialization_initiated=0.275146s, others=0.000713s)
I0417 13:52:54.324455 138211156707136 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.282105s (all_items=0.000018s, per_item={'items': '0.00001836'}, temp_paths=0.282087)
I0417 13:52:54.325248 138211156707136 event_tracking.py:125] [process=0] [async] Finished blocking save in 1.46 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_main_20260417_125726/linen_xpk_main_20260417_125726_10_shardy_true/checkpoints/0.
I0417 13:52:54.325579 138082275919616 async_checkpointer.py:76] [process=0][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-17 14:12:54.325543
I0417 13:52:54.341371 138211156707136 checkpoint_manager.py:1560] [process=0][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize
I0417 13:52:54.341642 138080673949440 async_checkpointer.py:280] [process=0][thread=save_finalize] Waiting for background save thread=async_save.
I0417 13:52:54.341797 138211156707136 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_main_20260417_125726/linen_xpk_main_20260417_125726_10_shardy_true/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776433972.859288, 'wait_for_prev_duration_secs': 5.793571472167969e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776433972.861358, 'checkpointer_blocking_duration_secs': 1.4643681049346924, 'get_old_steps_start_time': 1776433974.3257504, 'get_old_steps_duration_secs': 2.9325485229492188e-05, 'checkpoint_manager_blocking_start_time': 1776433972.8572972, 'checkpoint_manager_blocking_duration_secs': 1.4844646453857422}
I0417 13:52:54.341921 138211156707136 checkpointing.py:408] Started an asynchronous checkpoint save for step 0
I0417 13:52:54.341969 138211156707136 max_utils.py:750] 
Memstats: After params initialized:
I0417 13:52:54.342020 138211156707136 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_0(process=0,(0,0,0,0))
I0417 13:52:54.342052 138211156707136 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_1(process=0,(1,0,0,0))
I0417 13:52:54.342078 138211156707136 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_4(process=0,(0,1,0,0))
I0417 13:52:54.342102 138211156707136 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_5(process=0,(1,1,0,0))
I0417 13:52:54.679189 138211156707136 metric_logger.py:196] completed step: 0, seconds: 11.244, TFLOP/s/device: 1.208, Tokens/s/device: 182.134, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52779.875
I0417 13:52:54.682450 138211156707136 metric_logger.py:281] To see full metrics 'tensorboard --logdir=gs://lance-maxtext/linen_ckpt_xpk_main_20260417_125726/linen_xpk_main_20260417_125726_10_shardy_true/tensorboard/'
I0417 13:52:54.827649 138082267526912 atomicity.py:140] Creating tmp directory gs://lance-maxtext/linen_ckpt_xpk_main_20260417_125726/linen_xpk_main_20260417_125726_10_shardy_true/checkpoints/0
I0417 13:52:55.518410 138211156707136 metric_logger.py:196] completed step: 1, seconds: 1.974, TFLOP/s/device: 6.883, Tokens/s/device: 1037.496, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52779.875
I0417 13:52:55.664300 138211156707136 metric_logger.py:196] completed step: 2, seconds: 0.840, TFLOP/s/device: 16.177, Tokens/s/device: 2438.362, total_weights: 65536, loss: 10.262, lm_loss: 10.262, perplexity: 28637.129
I0417 13:52:55.811741 138211156707136 metric_logger.py:196] completed step: 3, seconds: 0.009, TFLOP/s/device: 1461.768, Tokens/s/device: 220333.513, total_weights: 65536, loss: 9.734, lm_loss: 9.734, perplexity: 16889.203
I0417 13:52:59.636914 138077464946432 checkpoint.py:188] Wrote Metadata={'item_handlers': None, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776433978715188031, 'commit_timestamp_nsecs': None, 'custom_metadata': {}}, json={"item_handlers": null, "metrics": {}, "performance_metrics": {}, "init_timestamp_nsecs": 1776433978715188031, "commit_timestamp_nsecs": null, "custom_metadata": {}} to gs://lance-maxtext/linen_ckpt_xpk_main_20260417_125726/linen_xpk_main_20260417_125726_10_shardy_true/checkpoints/0/_CHECKPOINT_METADATA
I0417 13:52:59.825088 138081780885248 atomicity.py:140] Creating tmp directory gs://lance-maxtext/linen_ckpt_xpk_main_20260417_125726/linen_xpk_main_20260417_125726_10_shardy_true/checkpoints/0/items
I0417 13:53:03.170012    2814 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com
I0417 13:53:07.130491 138081223042816 array_metadata_store.py:203] [process=0][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_main_20260417_125726/linen_xpk_main_20260417_125726_10_shardy_true/checkpoints/0/items/array_metadatas/process_0
I0417 13:53:22.615472 138211156707136 metric_logger.py:196] completed step: 4, seconds: 0.147, TFLOP/s/device: 92.426, Tokens/s/device: 13931.404, total_weights: 65536, loss: 9.277, lm_loss: 9.277, perplexity: 10694.614
I0417 13:53:22.624046 138211156707136 metric_logger.py:196] completed step: 5, seconds: 0.148, TFLOP/s/device: 92.027, Tokens/s/device: 13871.298, total_weights: 65536, loss: 8.892, lm_loss: 8.892, perplexity: 7270.717
I0417 13:53:22.770298 138211156707136 metric_logger.py:196] completed step: 6, seconds: 26.802, TFLOP/s/device: 0.507, Tokens/s/device: 76.413, total_weights: 65536, loss: 8.592, lm_loss: 8.592, perplexity: 5390.759
I0417 13:53:22.917699 138211156707136 metric_logger.py:196] completed step: 7, seconds: 0.006, TFLOP/s/device: 2102.294, Tokens/s/device: 316880.706, total_weights: 65536, loss: 8.384, lm_loss: 8.384, perplexity: 4376.905
I0417 13:53:23.065058 138211156707136 metric_logger.py:196] completed step: 8, seconds: 0.147, TFLOP/s/device: 92.357, Tokens/s/device: 13921.082, total_weights: 65536, loss: 8.255, lm_loss: 8.255, perplexity: 3846.406
I0417 13:53:23.211932 138211156707136 checkpointing.py:772] Waiting for step 9 to finish before checkpoint...
I0417 13:53:23.212605 138211156707136 checkpointing.py:776] Waited 0.0006945133209228516 seconds for step 9 to finish before starting checkpointing.
I0417 13:53:23.214745 138211156707136 checkpoint_manager.py:2020] [process=0][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0417 13:53:40.374744 138081214650112 base_pytree_checkpoint_handler.py:1282] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 37.204042s (commit=35.386879s, array_metadata_write=1.817163s)
I0417 13:53:40.375860 138082275919616 base_pytree_checkpoint_handler.py:130] [process=0] /jax/orbax/write/gbytes_per_sec: 34.095 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 46.3296115398407 s) (per-host)
I0417 13:53:40.375975 138082275919616 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 46.050287s.
I0417 13:53:42.444014 138082275919616 checkpoint.py:228] Read Metadata={'item_handlers': None, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776433978715188031, 'commit_timestamp_nsecs': None, 'custom_metadata': {}} from gs://lance-maxtext/linen_ckpt_xpk_main_20260417_125726/linen_xpk_main_20260417_125726_10_shardy_true/checkpoints/0/_CHECKPOINT_METADATA
I0417 13:53:43.644316 138077464946432 checkpoint.py:247] Updated Metadata={'item_handlers': {'items': 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler'}, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776433978715188031, 'commit_timestamp_nsecs': None, 'custom_metadata': {}} to gs://lance-maxtext/linen_ckpt_xpk_main_20260417_125726/linen_xpk_main_20260417_125726_10_shardy_true/checkpoints/0/_CHECKPOINT_METADATA
I0417 13:53:44.234356 138082275919616 ocdbt_utils.py:49] Param validation support for Zarr3 will be added later (b/362328389).
I0417 13:53:44.508764 138082275919616 array_metadata_store.py:411] [process=0][thread=async_save] Validated ArrayMetadata from all 8 hosts. Time taken: 0.000246s.
I0417 13:53:45.167703 138082275919616 base_pytree_checkpoint_handler.py:1406] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 2.597086s. use_zarr3=True, enable_post_merge_validation=True, directory=gs://lance-maxtext/linen_ckpt_xpk_main_20260417_125726/linen_xpk_main_20260417_125726_10_shardy_true/checkpoints/0/items
I0417 13:53:45.169658 138082275919616 atomicity.py:666] Finalizing gs://lance-maxtext/linen_ckpt_xpk_main_20260417_125726/linen_xpk_main_20260417_125726_10_shardy_true/checkpoints/0/items
I0417 13:53:46.566303 138082275919616 atomicity.py:666] Finalizing gs://lance-maxtext/linen_ckpt_xpk_main_20260417_125726/linen_xpk_main_20260417_125726_10_shardy_true/checkpoints/0
I0417 13:53:48.896754 138082275919616 atomicity.py:847] [process=0][thread=async_save] Finished saving checkpoint (finalized tmp dir) to `gs://lance-maxtext/linen_ckpt_xpk_main_20260417_125726/linen_xpk_main_20260417_125726_10_shardy_true/checkpoints/0`.
I0417 13:53:48.897638 138082275919616 event_tracking.py:138] [process=0] [async] Finished save (blocking + background) in 56.04 seconds @ gs://lance-maxtext/linen_ckpt_xpk_main_20260417_125726/linen_xpk_main_20260417_125726_10_shardy_true/checkpoints/0
I0417 13:53:48.899011 138082275919616 async_checkpointer.py:160] [process=0][thread=async_save] Background save thread done. Time taken: 54.573321s.
I0417 13:53:48.899237 138080673949440 async_checkpointer.py:288] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save.
I0417 13:53:48.899404 138080673949440 async_checkpointer.py:298] [process=0][thread=save_finalize] No errors found in background save thread=async_save.
I0417 13:53:48.899468 138080673949440 checkpoint_manager.py:2137] [process=0][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts...
I0417 13:53:48.902116 138080673949440 checkpoint_manager.py:2146] [process=0][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts.
I0417 13:53:48.902253 138211156707136 checkpoint_manager.py:2032] [process=0][thread=MainThread][step=0][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=0.
W0417 13:53:48.902340 138211156707136 checkpoint_manager.py:1452] Waiting for previous save to complete took 25.687599 seconds. If this number is high, consider checkpointing less frequently.
I0417 13:53:48.904257 138211156707136 checkpoint_manager.py:1512] [process=0] Saving checkpoint at step 9
I0417 13:53:48.906279 138211156707136 event_tracking.py:70] [process=0] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_main_20260417_125726/linen_xpk_main_20260417_125726_10_shardy_true/checkpoints/9.
I0417 13:53:49.702327 138211156707136 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array.
I0417 13:53:49.702423 138211156707136 replica_slices.py:424] Transferring arrays to host memory with options: use_replica_parallel=True, min_slice_bytes_for_replica_parallel=None, max_replicas_for_replica_parallel=None, enable_pinned_host_transfer=False
I0417 13:53:49.729124 138211156707136 base_pytree_checkpoint_handler.py:154] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.027524s
I0417 13:53:49.729990 138211156707136 base_pytree_checkpoint_handler.py:130] [process=0] /jax/orbax/write/blocking_gbytes_per_sec: 47.862 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.03222942352294922 s) (per-host)
I0417 13:53:49.730077 138211156707136 base_pytree_checkpoint_handler.py:768] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 0.032334s (batch_requests_ready=0.001986s, total_serialization_initiated=0.029567s, others=0.000781s)
I0417 13:53:49.730178 138211156707136 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.036499s (all_items=0.000019s, per_item={'items': '0.00001907'}, temp_paths=0.036480)
I0417 13:53:49.730958 138211156707136 event_tracking.py:125] [process=0] [async] Finished blocking save in 0.83 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_main_20260417_125726/linen_xpk_main_20260417_125726_10_shardy_true/checkpoints/9.
I0417 13:53:49.731270 138080665556736 async_checkpointer.py:76] [process=0][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-17 14:13:49.731239
I0417 13:53:49.739750 138211156707136 checkpoint_manager.py:1560] [process=0][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize
I0417 13:53:49.740051 138080673949440 async_checkpointer.py:280] [process=0][thread=save_finalize] Waiting for background save thread=async_save.
I0417 13:53:49.740217 138211156707136 standard_logger.py:34] {'step': 9, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_main_20260417_125726/linen_xpk_main_20260417_125726_10_shardy_true/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776434003.2147133, 'wait_for_prev_duration_secs': 25.687598943710327, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776434028.9042966, 'checkpointer_blocking_duration_secs': 0.827089786529541, 'get_old_steps_start_time': 1776434029.731409, 'get_old_steps_duration_secs': 3.361701965332031e-05, 'checkpoint_manager_blocking_start_time': 1776434003.2128305, 'checkpoint_manager_blocking_duration_secs': 26.527350664138794}
I0417 13:53:49.740332 138211156707136 checkpointing.py:408] Started an asynchronous checkpoint save for step 9
I0417 13:53:49.740376 138211156707136 checkpoint_manager.py:2020] [process=0][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0417 13:53:49.944618 138082275919616 atomicity.py:140] Creating tmp directory gs://lance-maxtext/linen_ckpt_xpk_main_20260417_125726/linen_xpk_main_20260417_125726_10_shardy_true/checkpoints/9
I0417 13:53:54.751682 138081235621632 atomicity.py:140] Creating tmp directory gs://lance-maxtext/linen_ckpt_xpk_main_20260417_125726/linen_xpk_main_20260417_125726_10_shardy_true/checkpoints/9/items
I0417 13:54:01.099427 138080695064320 array_metadata_store.py:203] [process=0][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_main_20260417_125726/linen_xpk_main_20260417_125726_10_shardy_true/checkpoints/9/items/array_metadatas/process_0
I0417 13:54:40.002274 138080686671616 base_pytree_checkpoint_handler.py:1282] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 42.423855s (commit=40.327034s, array_metadata_write=2.096821s)
I0417 13:54:40.003344 138080665556736 base_pytree_checkpoint_handler.py:130] [process=0] /jax/orbax/write/gbytes_per_sec: 31.400 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 50.305559158325195 s) (per-host)
I0417 13:54:40.003471 138080665556736 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 50.272105s.
I0417 13:54:42.791631 138080665556736 array_metadata_store.py:411] [process=0][thread=async_save] Validated ArrayMetadata from all 8 hosts. Time taken: 0.000216s.
I0417 13:54:42.832678 138080665556736 ocdbt_utils.py:49] Param validation support for Zarr3 will be added later (b/362328389).
I0417 13:54:43.766869 138080665556736 base_pytree_checkpoint_handler.py:1406] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 2.398599s. use_zarr3=True, enable_post_merge_validation=True, directory=gs://lance-maxtext/linen_ckpt_xpk_main_20260417_125726/linen_xpk_main_20260417_125726_10_shardy_true/checkpoints/9/items
I0417 13:54:43.768569 138080665556736 atomicity.py:666] Finalizing gs://lance-maxtext/linen_ckpt_xpk_main_20260417_125726/linen_xpk_main_20260417_125726_10_shardy_true/checkpoints/9/items
I0417 13:54:44.730521 138080665556736 atomicity.py:666] Finalizing gs://lance-maxtext/linen_ckpt_xpk_main_20260417_125726/linen_xpk_main_20260417_125726_10_shardy_true/checkpoints/9
I0417 13:54:47.024751 138080665556736 atomicity.py:847] [process=0][thread=async_save] Finished saving checkpoint (finalized tmp dir) to `gs://lance-maxtext/linen_ckpt_xpk_main_20260417_125726/linen_xpk_main_20260417_125726_10_shardy_true/checkpoints/9`.
I0417 13:54:47.025585 138080665556736 event_tracking.py:138] [process=0] [async] Finished save (blocking + background) in 58.12 seconds @ gs://lance-maxtext/linen_ckpt_xpk_main_20260417_125726/linen_xpk_main_20260417_125726_10_shardy_true/checkpoints/9
I0417 13:54:47.026955 138080665556736 async_checkpointer.py:160] [process=0][thread=async_save] Background save thread done. Time taken: 57.295587s.
I0417 13:54:47.027079 138080673949440 async_checkpointer.py:288] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save.
I0417 13:54:47.027142 138080673949440 async_checkpointer.py:298] [process=0][thread=save_finalize] No errors found in background save thread=async_save.
I0417 13:54:47.027197 138080673949440 checkpoint_manager.py:2137] [process=0][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts...
I0417 13:54:47.029026 138080673949440 checkpoint_manager.py:2146] [process=0][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts.
I0417 13:54:47.029202 138211156707136 checkpoint_manager.py:2032] [process=0][thread=MainThread][step=9][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=9.
I0417 13:54:47.029351 138211156707136 checkpoint_manager.py:2009] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0417 13:54:47.030433 138211156707136 metric_logger.py:196] completed step: 9, seconds: 0.148, TFLOP/s/device: 92.043, Tokens/s/device: 13873.741, total_weights: 65536, loss: 8.179, lm_loss: 8.179, perplexity: 3564.635
Per train step:
 Total TFLOPs: 13.59 
 split as 93.93% learnable weight flops and 6.07% attention flops
XPK End: Fri Apr 17 13:54:59 UTC 2026
EXIT_CODE=0
feat/nnx-post-train-fixes  ·  2fbc8c2f1  ·  feat_nnx_post_train_fixes_20260417_023710  ·  full log 
XPK Start: Fri Apr 17 03:22:24 UTC 2026
PyTorch was not found. Models won't be available and only tokenizers, configuration and file/data utilities can be used.
2026-04-17 03:22:48.447153: E external/local_xla/xla/stream_executor/cuda/cuda_platform.cc:51] failed call to cuInit: INTERNAL: CUDA error: Failed call to cuInit: UNKNOWN ERROR (303)
I0417 03:22:48.979272 138489075631936 max_utils.py:273] Attempting to initialize the jax distributed system...
I0417 03:22:58.019321 138489075631936 distributed.py:149] Starting JAX distributed service on [::]:8482
I0417 03:22:58.023874 138489075631936 distributed.py:172] Connecting to JAX distributed service on mt-10-shardy-true-rd5fn-slice-job-0-0.mt-10-shardy-true-rd5fn:8482
I0417 03:22:59.032659 138489075631936 max_utils.py:284] Jax distributed system initialized!
I0417 03:23:05.408089 138489075631936 max_utils.py:800] System Information: Jax Version: 0.9.2
I0417 03:23:05.408190 138489075631936 max_utils.py:801] System Information: Jaxlib Version: 0.9.2
I0417 03:23:05.408229 138489075631936 max_utils.py:802] System Information: Jax Backend: PJRT C API
TFRT TPU v6 lite
Built on Mar 4 2026 11:32:08 (1772652728) cl/878335365
I0417 03:23:05.408263 138489075631936 train_utils.py:377] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing.
I0417 03:23:06.072031 138489075631936 maxtext_utils.py:1687] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1)
I0417 03:23:06.072606 138489075631936 maxtext_utils.py:1687] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1)
I0417 03:23:06.072783 138489075631936 checkpointing.py:688] Setting up checkpoint logger...
I0417 03:23:06.072833 138489075631936 checkpointing.py:234] Creating checkpoint manager with ocdbt=True and zarr3=True
I0417 03:23:06.072896 138489075631936 pytree_checkpoint_handler.py:592] save_device_host_concurrent_bytes=None
I0417 03:23:06.073232 138489075631936 base_pytree_checkpoint_handler.py:441] Created BasePyTreeCheckpointHandler: use_ocdbt=True, use_zarr3=True, pytree_metadata_options=PyTreeMetadataOptions(support_rich_types=False), array_metadata_store=<orbax.checkpoint._src.metadata.array_metadata_store.Store object at 0x7df3c864c4a0>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB)
I0417 03:23:09.264362 138489075631936 checkpointing.py:266] Enabling policy for fixed interval checkpointing.
I0417 03:23:09.264593 138489075631936 checkpoint_manager.py:708] [process=0][thread=MainThread] CheckpointManager init: checkpointers=None, item_names=('items',), item_handlers={'items': <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7ddf9c7a5970>}, handler_registry=None
I0417 03:23:09.264831 138489075631936 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7ddf9c7a5970>` for item "items" and save args `<class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeSaveArgs'>` and restore args `<class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>` to `_handler_registry`.
I0417 03:23:09.264892 138489075631936 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7ddfb41fcbf0>` for item "metrics" and save args `<class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>` and restore args `<class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>` to `_handler_registry`.
I0417 03:23:09.264928 138489075631936 composite_checkpoint_handler.py:505] Initialized registry DefaultCheckpointHandlerRegistry({('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeSaveArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7ddf9c7a5970>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7ddf9c7a5970>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7ddfb41fcbf0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7ddfb41fcbf0>}).
I0417 03:23:09.265240 138489075631936 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.34
I0417 03:23:09.265308 138489075631936 async_checkpointer.py:192] [process=0][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>._fn at 0x7ddf9c3ec680> timeout: 1200 secs and primary_host=0 for async checkpoint writes
I0417 03:23:11.407537 138489075631936 checkpoint_manager.py:1812] Found 0 checkpoint steps in gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/linen_xpk_feat_nnx_post_train_fixes_20260417_023710_10_shardy_true/checkpoints
I0417 03:23:11.431183 138489075631936 checkpoint_manager.py:929] [process=0][thread=MainThread] CheckpointManager created,  primary_host=0, CheckpointManagerOptions=CheckpointManagerOptions(save_interval_steps=1, max_to_keep=None, keep_time_interval=None, keep_period=None, should_keep_fn=None, best_fn=None, best_mode='max', keep_checkpoints_without_metrics=True, step_prefix=None, step_format_fixed_length=None, step_name_format=None, create=True, cleanup_tmp_directories=False, save_on_steps=frozenset(), single_host_load_and_broadcast=False, todelete_subdir=None, todelete_full_path=None, enable_background_delete=False, read_only=False, enable_async_checkpointing=True, async_options=None, multiprocessing_options=MultiprocessingOptions(primary_host=0, active_processes=None, barrier_sync_key_prefix=None), should_save_fn=None, file_options=FileOptions(path_permission_mode=None), save_root_metadata=True, temporary_path_class=None, save_decision_policy=FixedIntervalPolicy(interval=10), preservation_policy=LatestN(n=None), prevent_write_metrics=False, enable_should_save_is_saving_in_progress_check=True, enable_per_process_directory_creation=False, lightweight_initialize=False), root_directory=gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/linen_xpk_feat_nnx_post_train_fixes_20260417_023710_10_shardy_true/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7ddf9c646450>
I0417 03:23:11.431318 138489075631936 checkpointing.py:302] Checkpoint manager created!
I0417 03:23:12.346746 138489075631936 nnx_wrappers.py:437] Unknown Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed').
I0417 03:23:12.346882 138489075631936 nnx_wrappers.py:437] Unknown Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0417 03:23:12.729696 138489075631936 attentions.py:1088] attentions/inputs_q Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed').
I0417 03:23:12.729789 138489075631936 attentions.py:1088] attentions/inputs_q Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0417 03:23:12.746370 138489075631936 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed').
I0417 03:23:12.746433 138489075631936 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0417 03:23:12.770319 138489075631936 attentions.py:1154] attentions/query Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0417 03:23:12.770396 138489075631936 attentions.py:1154] attentions/query Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0417 03:23:12.787196 138489075631936 attentions.py:1155] attentions/key Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0417 03:23:12.787262 138489075631936 attentions.py:1155] attentions/key Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0417 03:23:12.803948 138489075631936 attentions.py:1156] attentions/value Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0417 03:23:12.804012 138489075631936 attentions.py:1156] attentions/value Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0417 03:23:12.829679 138489075631936 attentions.py:1197] attentions/out Logical: bfloat16[32,2048,16,128].................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv').
I0417 03:23:12.829759 138489075631936 attentions.py:1197] attentions/out Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0417 03:23:12.850567 138489075631936 linears.py:525] linears/x Logical: bfloat16[32,2048,7168]...................................... ('activation_batch', 'activation_length', 'activation_mlp').
I0417 03:23:12.850637 138489075631936 linears.py:525] linears/x Physical: bfloat16[32,2048,7168]...................................... ('fsdp', None, None).
I0417 03:23:13.064759 138489075631936 checkpointing.py:578] checkpoint manager exists so trying to load this run's existing checkpoint
I0417 03:23:13.064888 138489075631936 checkpointing.py:676] No existing checkpoints found, not restoring checkpoint.
fsdp: 32
I0417 03:23:14.482206 138489075631936 maxtext_utils.py:1796]  params/params/decoder/decoder_norm/scale
    Shape:     float32[2048]
    Logical:   P('norm',)
    Physical:  (None,)
I0417 03:23:14.482358 138489075631936 maxtext_utils.py:1796]  params/params/decoder/layers/mlp/wi_0/kernel
    Shape:     float32[2048,16,7168]
    Logical:   P('embed', 'layers', 'mlp')
    Physical:  ('fsdp', None, None)
I0417 03:23:14.482434 138489075631936 maxtext_utils.py:1796]  params/params/decoder/layers/mlp/wi_1/kernel
    Shape:     float32[2048,16,7168]
    Logical:   P('embed', 'layers', 'mlp')
    Physical:  ('fsdp', None, None)
I0417 03:23:14.482524 138489075631936 maxtext_utils.py:1796]  params/params/decoder/layers/mlp/wo/kernel
    Shape:     float32[7168,16,2048]
    Logical:   P('mlp', 'layers', 'embed')
    Physical:  (None, None, 'fsdp')
I0417 03:23:14.482608 138489075631936 maxtext_utils.py:1796]  params/params/decoder/layers/post_self_attention_layer_norm/scale
    Shape:     float32[2048,16]
    Logical:   P('norm', 'layers')
    Physical:  (None, None)
I0417 03:23:14.482670 138489075631936 maxtext_utils.py:1796]  params/params/decoder/layers/pre_self_attention_layer_norm/scale
    Shape:     float32[2048,16]
    Logical:   P('norm', 'layers')
    Physical:  (None, None)
I0417 03:23:14.482749 138489075631936 maxtext_utils.py:1796]  params/params/decoder/layers/self_attention/key/kernel
    Shape:     float32[2048,16,16,128]
    Logical:   P('embed', 'layers', 'kv_heads', 'kv_head_dim')
    Physical:  ('fsdp', None, None, None)
I0417 03:23:14.482827 138489075631936 maxtext_utils.py:1796]  params/params/decoder/layers/self_attention/out/kernel
    Shape:     float32[16,16,128,2048]
    Logical:   P('heads', 'layers', 'kv', 'embed')
    Physical:  (None, None, None, 'fsdp')
I0417 03:23:14.482910 138489075631936 maxtext_utils.py:1796]  params/params/decoder/layers/self_attention/query/kernel
    Shape:     float32[2048,16,16,128]
    Logical:   P('embed', 'layers', 'q_heads', 'kv')
    Physical:  ('fsdp', None, None, None)
I0417 03:23:14.482973 138489075631936 maxtext_utils.py:1796]  params/params/decoder/layers/self_attention/value/kernel
    Shape:     float32[2048,16,16,128]
    Logical:   P('embed', 'layers', 'kv_heads', 'kv_head_dim')
    Physical:  ('fsdp', None, None, None)
I0417 03:23:14.483051 138489075631936 maxtext_utils.py:1796]  params/params/decoder/logits_dense/kernel
    Shape:     float32[2048,32000]
    Logical:   P('embed_vocab', 'vocab')
    Physical:  ('fsdp', None)
I0417 03:23:14.483113 138489075631936 maxtext_utils.py:1796]  params/params/token_embedder/embedding
    Shape:     float32[32000,2048]
    Logical:   P('vocab', 'embed_vocab')

    Physical:  (None, 'fsdp')
I0417 03:23:14.976217 138489075631936 train.py:157] train/xent Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length').
I0417 03:23:14.976303 138489075631936 train.py:157] train/xent Physical: float32[32,2048]............................................ ('fsdp', None).
I0417 03:23:14.992165 138489075631936 train.py:164] train/z_loss Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length').
I0417 03:23:14.992226 138489075631936 train.py:164] train/z_loss Physical: float32[32,2048]............................................ ('fsdp', None).
I0417 03:23:25.889453 138489075631936 max_utils.py:791] Total memory size: 1.5 GB, Output size: 0.4 GB, Temp size: 1.1 GB, Argument size: 0.4 GB, Host temp size: 0.0 GB.
I0417 03:23:25.896978 138489075631936 metric_logger.py:289] number parameters: 1.104 billion
I0417 03:23:37.604424 138489075631936 checkpointing.py:794] Waiting for step 0 to finish before checkpoint...
I0417 03:23:37.757522 138489075631936 checkpointing.py:798] Waited 0.15308022499084473 seconds for step 0 to finish before starting checkpointing.
I0417 03:23:37.760142 138489075631936 checkpoint_manager.py:2009] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0417 03:23:37.761948 138489075631936 checkpoint_manager.py:1512] [process=0] Saving checkpoint at step 0
I0417 03:23:37.763512 138489075631936 event_tracking.py:70] [process=0] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/linen_xpk_feat_nnx_post_train_fixes_20260417_023710_10_shardy_true/checkpoints/0.
I0417 03:23:38.095875 138489075631936 signaling_client.py:364] Using JaxDistributedSignalingClient
I0417 03:23:38.109964 138489075631936 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array.
I0417 03:23:38.110032 138489075631936 replica_slices.py:424] Transferring arrays to host memory with options: use_replica_parallel=True, min_slice_bytes_for_replica_parallel=None, max_replicas_for_replica_parallel=None, enable_pinned_host_transfer=False
I0417 03:23:38.361644 138360255170304 atomicity.py:140] Creating tmp directory gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/linen_xpk_feat_nnx_post_train_fixes_20260417_023710_10_shardy_true/checkpoints/0
I0417 03:23:38.376335 138489075631936 base_pytree_checkpoint_handler.py:154] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.267162s
I0417 03:23:38.377169 138489075631936 base_pytree_checkpoint_handler.py:130] [process=0] /jax/orbax/write/blocking_gbytes_per_sec: 5.640 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.2735011577606201 s) (per-host)
I0417 03:23:38.377247 138489075631936 base_pytree_checkpoint_handler.py:768] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 0.273605s (batch_requests_ready=0.002219s, total_serialization_initiated=0.270629s, others=0.000757s)
I0417 03:23:38.377327 138489075631936 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.277529s (all_items=0.000017s, per_item={'items': '0.00001740'}, temp_paths=0.277512)
I0417 03:23:38.378104 138489075631936 event_tracking.py:125] [process=0] [async] Finished blocking save in 0.62 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/linen_xpk_feat_nnx_post_train_fixes_20260417_023710_10_shardy_true/checkpoints/0.
I0417 03:23:38.378434 138360263563008 async_checkpointer.py:76] [process=0][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-17 03:43:38.378396
I0417 03:23:38.401337 138489075631936 checkpoint_manager.py:1560] [process=0][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize
I0417 03:23:38.401617 138357568952064 async_checkpointer.py:280] [process=0][thread=save_finalize] Waiting for background save thread=async_save.
I0417 03:23:38.401776 138489075631936 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/linen_xpk_feat_nnx_post_train_fixes_20260417_023710_10_shardy_true/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776396217.7601237, 'wait_for_prev_duration_secs': 6.175041198730469e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776396217.7619877, 'checkpointer_blocking_duration_secs': 0.6165969371795654, 'get_old_steps_start_time': 1776396218.378609, 'get_old_steps_duration_secs': 2.9325485229492188e-05, 'checkpoint_manager_blocking_start_time': 1776396217.758062, 'checkpoint_manager_blocking_duration_secs': 0.6436741352081299}
I0417 03:23:38.401902 138489075631936 checkpointing.py:409] Started an asynchronous checkpoint save for step 0
I0417 03:23:38.401954 138489075631936 max_utils.py:750] 
Memstats: After params initialized:
I0417 03:23:38.402007 138489075631936 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_0(process=0,(0,0,0,0))
I0417 03:23:38.402039 138489075631936 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_1(process=0,(1,0,0,0))
I0417 03:23:38.402065 138489075631936 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_4(process=0,(0,1,0,0))
I0417 03:23:38.402088 138489075631936 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_5(process=0,(1,1,0,0))
I0417 03:23:38.730301 138489075631936 metric_logger.py:185] completed step: 0, seconds: 11.409, TFLOP/s/device: 1.191, Tokens/s/device: 179.511, total_weights: 65536, loss: 10.874
I0417 03:23:38.733388 138489075631936 metric_logger.py:269] To see full metrics 'tensorboard --logdir=gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/linen_xpk_feat_nnx_post_train_fixes_20260417_023710_10_shardy_true/tensorboard/'
I0417 03:23:39.151620 138489075631936 metric_logger.py:185] completed step: 1, seconds: 1.124, TFLOP/s/device: 12.084, Tokens/s/device: 1821.464, total_weights: 65536, loss: 10.874
I0417 03:23:39.297563 138489075631936 metric_logger.py:185] completed step: 2, seconds: 0.422, TFLOP/s/device: 32.194, Tokens/s/device: 4852.621, total_weights: 65536, loss: 10.262
I0417 03:23:39.444999 138489075631936 metric_logger.py:185] completed step: 3, seconds: 0.009, TFLOP/s/device: 1560.304, Tokens/s/device: 235186.036, total_weights: 65536, loss: 9.734
I0417 03:23:40.570709 138359745799936 atomicity.py:140] Creating tmp directory gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/linen_xpk_feat_nnx_post_train_fixes_20260417_023710_10_shardy_true/checkpoints/0/items
I0417 03:23:40.725126 138355996096256 checkpoint.py:188] Wrote Metadata={'item_handlers': None, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776396220303119716, 'commit_timestamp_nsecs': None, 'custom_metadata': {}}, json={"item_handlers": null, "metrics": {}, "performance_metrics": {}, "init_timestamp_nsecs": 1776396220303119716, "commit_timestamp_nsecs": null, "custom_metadata": {}} to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/linen_xpk_feat_nnx_post_train_fixes_20260417_023710_10_shardy_true/checkpoints/0/_CHECKPOINT_METADATA
I0417 03:23:42.185090    2828 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com
I0417 03:23:44.111983 138358628898560 array_metadata_store.py:203] [process=0][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/linen_xpk_feat_nnx_post_train_fixes_20260417_023710_10_shardy_true/checkpoints/0/items/array_metadatas/process_0
I0417 03:24:01.768308 138489075631936 metric_logger.py:185] completed step: 4, seconds: 0.147, TFLOP/s/device: 92.518, Tokens/s/device: 13945.254, total_weights: 65536, loss: 9.277
I0417 03:24:01.776471 138489075631936 metric_logger.py:185] completed step: 5, seconds: 0.148, TFLOP/s/device: 92.093, Tokens/s/device: 13881.264, total_weights: 65536, loss: 8.892
I0417 03:24:01.922398 138489075631936 metric_logger.py:185] completed step: 6, seconds: 22.322, TFLOP/s/device: 0.609, Tokens/s/device: 91.749, total_weights: 65536, loss: 8.592
I0417 03:24:02.070392 138489075631936 metric_logger.py:185] completed step: 7, seconds: 0.006, TFLOP/s/device: 2239.883, Tokens/s/device: 337619.519, total_weights: 65536, loss: 8.384
I0417 03:24:02.217249 138489075631936 metric_logger.py:185] completed step: 8, seconds: 0.147, TFLOP/s/device: 92.384, Tokens/s/device: 13925.058, total_weights: 65536, loss: 8.255
I0417 03:24:02.364281 138489075631936 checkpointing.py:794] Waiting for step 9 to finish before checkpoint...
I0417 03:24:02.365002 138489075631936 checkpointing.py:798] Waited 0.0007424354553222656 seconds for step 9 to finish before starting checkpointing.
I0417 03:24:02.368196 138489075631936 checkpoint_manager.py:2020] [process=0][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0417 03:24:18.733796 138357577344768 base_pytree_checkpoint_handler.py:1282] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 36.557017s (commit=35.290038s, array_metadata_write=1.266979s)
I0417 03:24:18.736015 138360263563008 base_pytree_checkpoint_handler.py:130] [process=0] /jax/orbax/write/gbytes_per_sec: 38.875 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 40.632333755493164 s) (per-host)
I0417 03:24:18.736125 138360263563008 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 40.357582s.
I0417 03:24:19.482429 138360263563008 checkpoint.py:228] Read Metadata={'item_handlers': None, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776396220303119716, 'commit_timestamp_nsecs': None, 'custom_metadata': {}} from gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/linen_xpk_feat_nnx_post_train_fixes_20260417_023710_10_shardy_true/checkpoints/0/_CHECKPOINT_METADATA
I0417 03:24:20.917174 138360263563008 ocdbt_utils.py:49] Param validation support for Zarr3 will be added later (b/362328389).
I0417 03:24:20.967933 138360263563008 array_metadata_store.py:411] [process=0][thread=async_save] Validated ArrayMetadata from all 8 hosts. Time taken: 0.000263s.
I0417 03:24:21.091167 138355996096256 checkpoint.py:247] Updated Metadata={'item_handlers': {'items': 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler'}, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776396220303119716, 'commit_timestamp_nsecs': None, 'custom_metadata': {}} to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/linen_xpk_feat_nnx_post_train_fixes_20260417_023710_10_shardy_true/checkpoints/0/_CHECKPOINT_METADATA
I0417 03:24:21.549185 138360263563008 base_pytree_checkpoint_handler.py:1406] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 1.509131s. use_zarr3=True, enable_post_merge_validation=True, directory=gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/linen_xpk_feat_nnx_post_train_fixes_20260417_023710_10_shardy_true/checkpoints/0/items
I0417 03:24:21.551388 138360263563008 atomicity.py:666] Finalizing gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/linen_xpk_feat_nnx_post_train_fixes_20260417_023710_10_shardy_true/checkpoints/0/items
I0417 03:24:22.503175 138360263563008 atomicity.py:666] Finalizing gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/linen_xpk_feat_nnx_post_train_fixes_20260417_023710_10_shardy_true/checkpoints/0
I0417 03:24:23.929714 138360263563008 atomicity.py:847] [process=0][thread=async_save] Finished saving checkpoint (finalized tmp dir) to `gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/linen_xpk_feat_nnx_post_train_fixes_20260417_023710_10_shardy_true/checkpoints/0`.
I0417 03:24:23.930574 138360263563008 event_tracking.py:138] [process=0] [async] Finished save (blocking + background) in 46.17 seconds @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/linen_xpk_feat_nnx_post_train_fixes_20260417_023710_10_shardy_true/checkpoints/0
I0417 03:24:23.931883 138360263563008 async_checkpointer.py:160] [process=0][thread=async_save] Background save thread done. Time taken: 45.553336s.
I0417 03:24:23.932077 138357568952064 async_checkpointer.py:288] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save.
I0417 03:24:23.932203 138357568952064 async_checkpointer.py:298] [process=0][thread=save_finalize] No errors found in background save thread=async_save.
I0417 03:24:23.932264 138357568952064 checkpoint_manager.py:2137] [process=0][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts...
I0417 03:24:23.935088 138357568952064 checkpoint_manager.py:2146] [process=0][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts.
I0417 03:24:23.935208 138489075631936 checkpoint_manager.py:2032] [process=0][thread=MainThread][step=0][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=0.
W0417 03:24:23.935288 138489075631936 checkpoint_manager.py:1452] Waiting for previous save to complete took 21.567098 seconds. If this number is high, consider checkpointing less frequently.
I0417 03:24:23.937307 138489075631936 checkpoint_manager.py:1512] [process=0] Saving checkpoint at step 9
I0417 03:24:23.939280 138489075631936 event_tracking.py:70] [process=0] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/linen_xpk_feat_nnx_post_train_fixes_20260417_023710_10_shardy_true/checkpoints/9.
I0417 03:24:24.654269 138489075631936 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array.
I0417 03:24:24.654357 138489075631936 replica_slices.py:424] Transferring arrays to host memory with options: use_replica_parallel=True, min_slice_bytes_for_replica_parallel=None, max_replicas_for_replica_parallel=None, enable_pinned_host_transfer=False
I0417 03:24:24.681657 138489075631936 base_pytree_checkpoint_handler.py:154] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.028163s
I0417 03:24:24.682472 138489075631936 base_pytree_checkpoint_handler.py:130] [process=0] /jax/orbax/write/blocking_gbytes_per_sec: 46.496 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.033176422119140625 s) (per-host)
I0417 03:24:24.682544 138489075631936 base_pytree_checkpoint_handler.py:768] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 0.033265s (batch_requests_ready=0.001994s, total_serialization_initiated=0.030514s, others=0.000757s)
I0417 03:24:24.682621 138489075631936 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.037353s (all_items=0.000016s, per_item={'items': '0.00001574'}, temp_paths=0.037337)
I0417 03:24:24.683325 138489075631936 event_tracking.py:125] [process=0] [async] Finished blocking save in 0.75 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/linen_xpk_feat_nnx_post_train_fixes_20260417_023710_10_shardy_true/checkpoints/9.
I0417 03:24:24.683611 138357568952064 async_checkpointer.py:76] [process=0][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-17 03:44:24.683579
I0417 03:24:24.690648 138489075631936 checkpoint_manager.py:1560] [process=0][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize
I0417 03:24:24.690897 138359183750912 async_checkpointer.py:280] [process=0][thread=save_finalize] Waiting for background save thread=async_save.
I0417 03:24:24.691032 138489075631936 standard_logger.py:34] {'step': 9, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/linen_xpk_feat_nnx_post_train_fixes_20260417_023710_10_shardy_true/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776396242.3681648, 'wait_for_prev_duration_secs': 21.567097663879395, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776396263.9373467, 'checkpointer_blocking_duration_secs': 0.7463967800140381, 'get_old_steps_start_time': 1776396264.6837666, 'get_old_steps_duration_secs': 3.0994415283203125e-05, 'checkpoint_manager_blocking_start_time': 1776396242.365243, 'checkpoint_manager_blocking_duration_secs': 22.325754404067993}
I0417 03:24:24.691142 138489075631936 checkpointing.py:409] Started an asynchronous checkpoint save for step 9
I0417 03:24:24.691189 138489075631936 checkpoint_manager.py:2020] [process=0][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0417 03:24:24.904758 138360263563008 atomicity.py:140] Creating tmp directory gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/linen_xpk_feat_nnx_post_train_fixes_20260417_023710_10_shardy_true/checkpoints/9
I0417 03:24:26.638575 138337826346752 atomicity.py:140] Creating tmp directory gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/linen_xpk_feat_nnx_post_train_fixes_20260417_023710_10_shardy_true/checkpoints/9/items
I0417 03:24:29.947089 138359200536320 array_metadata_store.py:203] [process=0][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/linen_xpk_feat_nnx_post_train_fixes_20260417_023710_10_shardy_true/checkpoints/9/items/array_metadatas/process_0
I0417 03:25:08.217986 138359192143616 base_pytree_checkpoint_handler.py:1282] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 40.465410s (commit=39.144559s, array_metadata_write=1.320851s)
I0417 03:25:08.219061 138357568952064 base_pytree_checkpoint_handler.py:130] [process=0] /jax/orbax/write/gbytes_per_sec: 36.254 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 43.569735527038574 s) (per-host)
I0417 03:25:08.219181 138357568952064 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 43.535461s.
I0417 03:25:10.429720 138357568952064 ocdbt_utils.py:49] Param validation support for Zarr3 will be added later (b/362328389).
I0417 03:25:10.532435 138357568952064 array_metadata_store.py:411] [process=0][thread=async_save] Validated ArrayMetadata from all 8 hosts. Time taken: 0.000231s.
I0417 03:25:11.074086 138357568952064 base_pytree_checkpoint_handler.py:1406] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 1.543645s. use_zarr3=True, enable_post_merge_validation=True, directory=gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/linen_xpk_feat_nnx_post_train_fixes_20260417_023710_10_shardy_true/checkpoints/9/items
I0417 03:25:11.075685 138357568952064 atomicity.py:666] Finalizing gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/linen_xpk_feat_nnx_post_train_fixes_20260417_023710_10_shardy_true/checkpoints/9/items
I0417 03:25:11.639084 138357568952064 atomicity.py:666] Finalizing gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/linen_xpk_feat_nnx_post_train_fixes_20260417_023710_10_shardy_true/checkpoints/9
I0417 03:25:13.076242 138357568952064 atomicity.py:847] [process=0][thread=async_save] Finished saving checkpoint (finalized tmp dir) to `gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/linen_xpk_feat_nnx_post_train_fixes_20260417_023710_10_shardy_true/checkpoints/9`.
I0417 03:25:13.077067 138357568952064 event_tracking.py:138] [process=0] [async] Finished save (blocking + background) in 49.14 seconds @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/linen_xpk_feat_nnx_post_train_fixes_20260417_023710_10_shardy_true/checkpoints/9
I0417 03:25:13.078511 138357568952064 async_checkpointer.py:160] [process=0][thread=async_save] Background save thread done. Time taken: 48.394788s.
I0417 03:25:13.078724 138359183750912 async_checkpointer.py:288] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save.
I0417 03:25:13.078854 138359183750912 async_checkpointer.py:298] [process=0][thread=save_finalize] No errors found in background save thread=async_save.
I0417 03:25:13.078926 138359183750912 checkpoint_manager.py:2137] [process=0][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts...
I0417 03:25:13.080741 138359183750912 checkpoint_manager.py:2146] [process=0][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts.
I0417 03:25:13.080918 138489075631936 checkpoint_manager.py:2032] [process=0][thread=MainThread][step=9][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=9.
I0417 03:25:13.081058 138489075631936 checkpoint_manager.py:2009] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0417 03:25:13.081690 138489075631936 metric_logger.py:185] completed step: 9, seconds: 0.150, TFLOP/s/device: 90.827, Tokens/s/device: 13690.480, total_weights: 65536, loss: 8.179
Per train step:
 Total TFLOPs: 13.59 
 split as 93.93% learnable weight flops and 6.07% attention flops
XPK End: Fri Apr 17 03:25:25 UTC 2026
EXIT_CODE=0